UBI517 Expert Systems 1 Lecture 2 Knowledge Representation

UBI517 Expert Systems 1

Lecture 2

Knowledge Representation

UBI 517 Expert Systems2

Knowledge representation

Epistemology or theory of knowledge is the branch of philosophy that studies the nature, methods, limitations, and validity of knowledge and belief.

In other words, epistemology primarily addresses the following questions: "What is knowledge?", "How is knowledge acquired?", and "What do people know?"

Wikipedia


Knowledge representation

Epistemology Addresses

The structure used to describe the elements of knowledge.

The interpretive process required to use the described knowledge.


Representation Criteria

Transparency Explicitness Naturalness Efficiency Adequacy Modularity


Knowledge Representation Schemes

Mypoulos and Lavesque (1984) classification: 1) Logical representation schemes

First order predicate logic for knowledge representation (scheme)Prolog for implementing (media)

2) Procedural representation schemes

Production systems


Knowledge Representation Schemes

3) Network representation schemesCapture knowledge as a graph in which the nodes represent the objects or concepts in the problem domain and the arcs represent the relations or associations between them. Semantic networks Conceptual graphs

4) Structured representation schemes

Each node is considered as a complex data structure consisting of named slots with attached values Frame-based systems


Formal Logic Logic provides

A representation of knowledge & Automation of the inferencing

process

Formal Logic• Propositional Logic• Predicate Logic


Propositional Logic

One of the oldest & simplest type of formal logic

Formal indicates logic is concerned with form of logical statements as opposed to meaning


Propositional Logic Propositional symbols denote propositions or

statements about the world that may be either true or false.

ExamplesIt is raining.

raining It is sunny.

sunny If it is raining then it is not sunny.

raining sunny


Propositional Logic

The truth table for the implication connective is shown below.


Propositional Logic

For propositional expressions P, Q, and R: P) = P (PQ) = (PQ) De Morgan's Law: (PQ) = (PQ) De Morgan's Law: (PQ) = (PQ) distributive law: PQR) = (PQ)PR) distributive law: PQR) = (PQ)PR) commutative law: (PQ) = (QP) commutative law: (PQ) = (QP) associative law: PQR) = (PQ) R associative law: PQR) = (PQ)R contrapositive law: (PQ) = (QP)


Propositional Logic

PREMISE: All X are YPREMISE: Z is a XConclusion: Z is a Y

Separating form from meaning is what gives logic its power as a tool

Propositional logic is a symbolic logic for manipulating propositions or logical variables

Propositions are classified as either T or F. KBS course is brilliant - (definitely T) Fish is good to eat - ( uncertain truth as depends on

person)

Doesn’t matter if X is apples, planes or Ships.Only the form is important.


Propositional Logic

He is tall - (variable involved so can’t assign value. Also tall is hard concept to define. Can’t be done in pred. Calculus).

Proposition logic: assign a truth value to statements Propositional logic provides a mechanism for

assigning a truth value to compound proposition based on value of individual propositions and connective involved

Propositional logic connectivesConjunction ANDDisjunction ORNegation NOT A’ Material implication If-Then -->Material equivalence Equals


Propositional Logic

Prop Logic can only deal with complete

sentences that is, it can not examine the internal structure of a statement.

too simple for complex domains no support for inferencing doesn’t handle fuzzy concepts

Solution - Predicate Logic


First order predicate logic

Predicate logic was developed in order to analyze more general cases. Propositional logic is a subset of predicate logic.

Concerned with internal structure of sentences

Quantifiers - all, some, no - make sentence more exact.

Wider scope of expression.


First order predicate logic

an alphabeta formal language axiomsinference rules


an alphabet

predicates represent the relations within the domain: man(tom)

variables - Dog, Color functions - father(CHILD), plus (2, 3) constants - rover(a_dog), blue(a_color) connectives - and, or, not, is equivalent

to.. quantifiers -X, X :

x [ PERSON(X) NEED-AIR(X)] delimiters


Language

A legally constructed formula in the language is called a well-formed formula (wff)

Evaluation is done by the use of truth tables.


Language

Knowledge engineering When Using Formal Logic

Develop an understanding of the knowledge.

Formulate the knowledge as English statements.

Break the statements into their component parts.

Choose symbols to represent the elements and relations in each component.

Build wffs by using the above symbols that represent the statement.

Predicate relation is derived by its name and its arity.


LanguageExample:

"Rover is a black dog“

"You will gain weight unless you exercise.“

“If the student’s average grade is greater than 90%, then the student will get an A in the course.”

black(rover) dog(rover)

X [¬exercise(X) gain_weight(X) ]

X, Name [(student(Name) avrg_grade(Name, X) ge(X, 90) final_grade(Name, “A”) ]


Inference ProcessInference in formal logic is the process of generating new wffs from existing wffs through the application of rules of inference.


Inference ProcessConclusion: Many English sentences are ambiguous. There is often a choice of ways of

representing the knowledge. Even in very simple situations a set of

sentences is unlikely to contain all the information necessary to reason about the topic at hand.


Inference ProcessLogical InferencesModus ponens and modus tolens provide

the foundation for making references.

Modus ponens: ((p q) p) qIf someone is snorkeling then he is wet

x snorkeling(X) wet(X)If we are given that alex is snorkeling

snorkeling (alex)we can infer wet(alex)


Inference ProcessLogical Inferences

Modus tolens: ((p q) q) p

If someone is snorkeling then he is wet

x snorkeling(X) wet(X)

If we are given that alex is not wet

wet(alex)

we can infer snorkeling(alex)


Inference ProcessThere are three reasoning methods that can

be applied to a set of premises.1. Deduction

2. Abduction

3. Induction


Inference ProcessDeduction is reasoning from known (premises) to

unknown (logical conclusions).x, y, z larger(x, y) larger(y, z) larger(x, z)

If our list of axioms contain the axioms

larger(house, car)

larger(car, cat)

Through deductive reasoning the wff

larger(house, cat)

Can be derived and added to our list of axioms


Inference ProcessIn Abduction we begin with a conclusion and

procede to derive conditions that would make the conclusion valid. In other words we try to find an explanation for the conclusion

It does not guarantee that a true conclusion results.

Therefore it s called unsound rule of inference.

If given B and B is trueAbduction allows us to say A is possibly true.

Reasoning under uncertainty.


Inference ProcessInduction Inductive reasoning forms the basis of scientific discovery.If given p(a) is true

p(b) is true ……..Then we conclude x, p(x) is true

If we observe alex over a period of time and note that whenever he is wet, it turns out that he has gone snorkeling.

We might induce that x, wet(x) snorkeling(x)

Like abduction, induction is also an unsound inference method.


Inference ProcessMonotonic vs non-monotonic reasoning Deductive reasoning is a monotonic reasoning that produce

as arguments that preserve truth.Axioms are not allowed to change, since once a fact is known

to be true, it is always true and can never be modified or retracted.

Most real life problems are non-monotonicquarter(fourth)leading(bucks)

Team [leading(Team) quarter(fourth)] strategy(Team, conservative)

We can deduce strategy(bucks, conservative)What if the state changes to

leading(dolphins)


Inference ProcessRESOLUTION attempts to show that the negation

of the statement produces a contradiction with the known statements.

winter V summer

~winter V cold deduce

summer V cold

In the above example if it is winter the first statement is true if not the second statement is true. From these two we can deduce the third statement to be true


Inference ProcessConjuctive normal form(Davis, 1960)

The steps to convert to conjuctive normal form: 1. Eliminate by using the fact that a b is

equivalent to ~a V b2. Reduce the scope of ~ ~(A V B) = ~A B~(A B) = ~A V B

DeMorgan's law ~x P(x) = x ~P(x) ~x P(x) = x ~P(x)


Inference Process3. Standardize variables so that each

quantifier binds a unique variable.

x P(x) V x Q(x) would be converted to

x P(x) V y Q(y)

4. Move all quantifiers to the left of the formula without changing their relative order.

x y P(x) V Q(y)


Inference Process5. Eliminate existential quantifiers. A

formula that contains an existentially quantified variable asserts that there is a value that can be substituted for the variable that makes the formula true.

y President(y) can be transformed into President(S1)


Inference Process5. If existential quantifiers occur within the scope of

a universal quantifier then the value that satisfies the predicate may depend on the values of the universally quantified variables.

x y fatherof(y, x) can be transformed into x fatherof(S2(x), x)

These generated functions are called Skolem functions.


Inference Process6. Drop the prefix.

7. Convert the matrix into conjunction of disjuncts.

8. Call each conjunct a separate clause.


Inference ProcessThe Unification Algorithm:

The object of the unification procedure is to discover at least one substitution that causes two literals to match.


Inference ProcessExample: 1. Marcus was a man.

man(marcus)2. Marcus was a Pompeian.

pompeian(marcus) 3. All Pompeians were Romans.

x pompeian(X) roman(X) 4. Caesar is a ruler.

ruler(caesar)


Inference Process5. All Romans were either loyal to Caesar or hated him. x roman(X) loyalto(X, casear) v hate(X, casear) 6. Everyone is loyal to someone.

X Y loyalto(X, Y) 7. People only try to assassinate rulers they are not loyal to.

X Y person(X) ruler(Y) tryassassinate(X,Y) ~loyalto(X, Y) 8. Marcus tried to assassinate Caesar.

tryassassinate(marcus, caesar)


Inference Process Proof by resolution: Given the axioms in clause form:1. man(Marcus)2. Pompeian(Marcus)3. ~Pompeian(x1) V Roman(x1)4. ruler(Caesar)5. ~Roman(x2) V loyalto(x2, Caesar) V hate(x2,

Caesar)6. loyalto(x3, f1(x3))7. ~person(x4) V ~ruler(y1) V ~tryassassinate(x4,

y1) V ~loyalto(x4, y1)8. tryassassinate(Marcus, Caesar)


Inference ProcessProve: hate(Marcus, Caesar)


Inference ProcessProve: loyalto(Marcus, Caesar)


Inference ProcessProve: loyalto(Marcus, Caesar) - Continue

Suppose our knowledge base contained the two additional statements

9. persecute(x, y) hate(y, x) ~persecute(x5, y2) V hate(y2, x5)

10. hate(x, y) persecute(y, x) ~hate(x6, y3) V persecute(y3, x6)


PrologProlog is the implementation language

for predicate logic.

Prolog is declarative

Three main features are:FactsRulesBacktracking

If something is not explicitly stated, Prolog assumes that it is false.

Prolog is non-monotonic where it deviates from pure predicate logic.


Examplechild_of(alex, nicole).child_of(alina, nicole).child_of(nicholas, leah).child_of(philip, leah).child_of(melanie, cathy).male(alex).male(philip).male(alex).male(nicholas).female(alina).female(leah).female(nicole).female(angela).sisters(nicole, leah).sisters(X, Z):- child_of(X, Y), child_of(Z, Y),

female(X), female(Z)


Disadvantages of Predicate Logic as a Basis for Knowledge-Based System

Managing uncertainty availability of only two levels of truth

– true or false. certainty factors have been

implemented in Prolog to mitigate this.

Monotonic versus non-monotonic reasoning.


Exercises:

Consider the following sentences: John likes all kinds of food. Apples are food. Chicken is food. Anything anyone eats and isn't killed by is food. Bill eats peanuts and is still alive. Sue eats everything Bill eats.

a. Translate these sentences into formulas in predicate logic.b. Convert the formulas of part a into clause form. c. Prove that John likes peanuts.d. Use resolution to answer the question, "What food does Sue

eat?"


Exercises:

Consider the following logic statements:x [panther(X) feline(X)]

x [house_cat(X) feline(X)]

x [house_cat(X) docile(X)]

x [feline(X) carnivore(X)]

x [carnivore(X) food(X, meat)]

x [horse(X) herbivore(X)]

x [herbivore(X) food(X, plants)]


Exercises:

a)Using modus ponens derive all possible relations from the following facts.panther(sam)house_cat(rubble)lion(leo)

b)Using abduction and the following statement what could kitty be?food(kitty, meat)

c)Convert all of the logic statements into clausesd) Attempt to prove the following statements using resolution:

house_cat(rebel) food(rebel, meat)horse(wilber) food(wilber, meat)

Documents

UBI517 Expert Systems 1 Lecture 2 Knowledge Representation